Fig. 5.

Model fit assessment and robustness of finding. (A) Our model specifying CV2 as a function of sex, age and genotype fits the observed data well. The blue line shows a LOESS-smoother for visual guidance: A good model fit requires that the residuals (y-axis) are independent of the CV2-values (x-axis) and generally lack ‘pattern’, which is the case here. The grouping of points into eight vertical bands is a consequence of the model being restricted to include only sex, age (two groups) and genotype, resulting in 2×2×2=8 possible combinations of covariate-values and, therefore, only eight distinct predicted (fitted) CV2 values. (B) A statistically significant association of genotype and CV2 is present in essentially all of the 10,000 bootstrap replications (98.28%) and is a robust feature of the data. Each bootstrap replication is based on re-sampling with replacement from the truly observed data with identical number of animals (but some possibly being included in the data several times due to the sampling with replacement approach while some will be dropped). The generally small P-values resulting from each bootstrap replication indicate that the statistical significance is not driven by a few extreme observations, but is a general feature of this data set.